BRPI0507229B1 - EQUIPMENT AND METHOD FOR DETERMINING THE QUANTIZATION STEP DIMENSION - Google Patents

Description

(54) Title: EQUIPMENT AND METHOD FOR DETERMINING THE DIMENSION OF THE STAGE OF

QUANTIZATION (51) lnt.CI .: G10L 19/02 (30) Unionist Priority: 01/03/2004 DE 10 2004 009 955.3 (73) Holder (s): FRAUNHOFER-GESELLSCHAFT ZUR FÕRDERUNG DER ANGEWANDTEN FORSCHUNG E.V.

(72) Inventor (s): BERNHARD GRILL; MICHAEL SCHUG; BODO TEICHMANN; NIKOLAUS RETTELBACH (85) National Phase Start Date: 08/24/2006

EQUIPMENT ·· ·

AND METHOD ·

FOR ’· Í UETEÉÍfflJÂVJÃÒ’ DA

DIMENSION OF THE QUANTIZATION STAGE

description

The present invention relates to audio encoders, and particularly to audio encoders that are based on transformation, i.e., characterized by the fact that a conversion of a temporal representation to a spectral representation is carried out at the beginning of the encoder pipeline.

An audio encoder of the prior art 10 based on transformation is shown in Fig. 3. The encoder shown in Fig. 3 is represented in the international standard ISO / IEC 14496-3: 2001 (Ε), subpart 4, page 4, being also known in the art as AAC encoder.

The prior art audio encoder will be shown below. An audio signal to be encoded is sent to an input 1000. This audio signal is initially sent to a graduation stage 1002, characterized by the fact that the gain control of said AAC is done to establish the level of the audio signal. Collateral information from graduation is

0 supplied to a bit stream formatter 1004, as represented by the arrow located between block 1002 and block 1004. The graduated audio signal is then sent to an MDCT filter bank 1006.

With the AAC encoder, the filter bank implements a modified discrete cosine transformation with 50% overlapping windows, with the window length being determined by a 1008 block.

In general, block 1008 exists with the objective of winding transient signals with relatively short windows, and fl • ·· ···· ·· ···· · • ·· ·! · • ····· «· • · · · · · · · · · ·· · · · · · · · · ········· ·· window signs that tend to be etêtáciõnários with windows relatively large. This serves to achieve a higher level of resolution time (at the expense of frequency resolution) for transient signals, due to the relatively short windows, characterized by the fact that for signals that tend to be stationary, a higher frequency resolution is achieved (at the expense of time resolution), due to longer windows, there is a tendency for the preference of longer windows, since they result in a greater coding gain. At the output of the filter bank 1006, there are blocks of spectral values - the blocks being successive in time - which can be MDCT coefficients, Fourier coefficients or subband signals, depending on the implementation of the filter bank, each sub-signal band having a specific limited bandwidth by the respective subband channel in filter bank 1006, and each subband signal having a specific number of subband samples.

This is followed by a presentation, by way of example, of the case in which the filter bank provides temporally successive blocks of MDCT spectral coefficients which, generally speaking, represent successive short-term spectra of audio signals to be encoded at input 1000. An MDCT spectral value block is then sent to a processing block

TNS 1010 (TNS = temporal noise modulation), where the temporal noise modulation is performed. The TNS technique is used to conform the temporal form of the quantization noise within each window of the transformation. This is done by applying a filtration process to the spectral data pieces for each channel. Coding is done on a window basis. In particular, the following steps are

- 3 :::: :: ..:. · ·

... ·· j · j:: * * _aa performed to apply the tool. TNS? * áftia ** window * * cLe spectral data, that is, a block of spectral values.

Initially, a frequency range for the TNS tool is selected. A suitable selection includes covering a frequency range of 1.5 kHz with a filter, up to the largest possible range of scale factor. It will be emphasized that this frequency range depends on the sampling rate, as specified in the AAC standard (ISO / IEC 14496-3: 2001 (E)).

Then, an LPC calculation is performed (LPC = linear predictive coding), precisely using the spectral MDCT coefficients existing in the selected range of target frequencies. For greater stability, coefficients that correspond to frequencies below 2.5 kHz are excluded from this process. Common LPC procedures as known in the speech process for LPC calculation can be used, for example, the well-known Levinson-Durbin algorithm. The calculation is made for the maximum permissible order of noise modulation filter.

As a result of the LPC calculation, the expected PG of prediction gain is obtained. In addition, reflection coefficients, or Parcor coefficients, are obtained.

If the prediction gain does not exceed a specific limit, the TNS tool does not apply. In this case, a piece of control information is written to the bit stream, so that the decoder knows that no TNS processing has been done.

However, if the prediction gain exceeds a threshold, TNS processing is applied.

In the next step, the reflection coefficients are quantized. The order of the noise modulation filter used is t

::: - .:: f /. · * ·· • * · * · · · · · * ****** determined by removing all ccfefiíVerttes' cie * rêíYêxao that have an absolute value below a limit from the tail of the set of reflection coefficients. The number of reflection coefficients remaining is on the order of magnitude of the noise modulation filter. An appropriate limit is 0.1.

The remaining reflection coefficients are typically converted into linear prediction coefficients, and this technique is also known as the step-up procedure.

The calculated LPC coefficients are then used as noise modulation filter coding coefficients, that is, as prediction filter coefficients. This FIR filter is used for filtering in the specified target frequency range. An auto-regressive filter is used in decoding, characterized by the fact that the referred average movement filter is used in the encoding. Eventually, collateral information for the TNS tool is provided to the bit stream formatter, as represented by the arrow shown between the TNS 1010 processor block and bit stream formatter 1004 in Fig.3.

Then, several optional tools that are not shown in Fig. 3 are traversed, such as a long-term prediction tool, an intensity / coupling tool, a prediction tool, a noise replacement tool, until eventually an encoder is reached center / side 1012. The center / side encoder 1012 will be active when the audio signal to be encoded is a multi-channel signal, that is, a stereo signal with a left channel and a right channel. Until now, that is, upstream of block 1012 of Fig. 3, the left stereo channels have been processed

Go· · ♦ · · ·· • * ·· and law, that is, graduated, transfcffcmadòb * £ eíõ Bank * ã * Filters, submitted or not to TNS processing, etc., separately between

In the central / lateral encoder, a check is initially made as to whether a central / lateral encoding makes sense, that is, whether it will provide an encoding gain. The central / lateral coding will produce a coding gain if the left and right channels tend to be similar, since in this case, the middle channel, that is, the sum of the left and right channels is almost equal to the left or right channel. , regardless of graduation by a factor of 1/2, characterized by the fact that the lateral channel has only very small values, since it is equal to the difference between the left and right channels. As a consequence, it can be seen that when the left and right channels are approximately the same, the difference will be approximately zero, or include only very small values that - and this is the hope - will be quantized to zero in a subsequent quantizer 1014, and so they can be transmitted very efficiently, since an entropy encoder 1016 is connected downstream of the quantizer 1014.

quantizer 1014 provides permissible interference per band of scale factor by a psychoacoustic model 1020. The quantizer operates in an iterative manner, that is, it is initially calling an external iteration circuit, which will then call an internal iteration circuit. In general, starting from scaled starting values, a quantization of a block of values is initially performed at the input of quantizer 1014. In particular, the internal circuit quantifies the «» «· · * * * * · ·» «·· · · * · *’ _ MDCT coefficients, with Hô ** process * üm ° * being consumed * specific number of bits. The external circuit calculates the distortion and the modified energy of the coefficients that use the scale factor, in order to recall an internal circuit. This process is iterated over a period of time until a specific conditional clause is reached. For each iteration in the external iteration circuit, the signal is reconstructed in order to calculate the interference introduced by the quantization, and to compare it with the permissible interference provided by the psycho-acoustic model

1020. In addition, the scale factors of these frequency bands, which after this comparison are still considered to be interfered, are increased by one or more iteration stages for iteration, to be exact, for each iteration of the external iteration circuit.

Once a situation is reached in which the quantization interference introduced by the quantization is less than the permissible interference determined by the psychoacoustic model, and if at the same time the bit requirements are obeyed, a state in which, to be exact, a maximum bit rate is not exceeded, the iteration, that is, the method of analysis by synthesis is completed, and the scale factors obtained are coded as illustrated in block 1014, being supplied in coded form to a streatn bit formatter 1004 as marked by the existing arrow between block 1014 and block 1004. The values

The quantized 5 is then sent to the entropy encoder 1016, which typically performs the entropy coding for various bands of scale factor using several code tables.

Huffman, in order to translate the quantized values into a

V • ·· 4 *

a »···» 4 ···· ** · ** · * '.. *, * · «· *. ! ! ϊ *. ·· «· · * · * · * · · · *.

- * 4 4 · · * Φ * ⁴ '9 »· * · *? · <

binary format. As it is known, • codíK.dãçãoõ ê entropy in the form of Huffman coding involves returning to coding tables that are created based on expected signal statistics, and where frequently occurring values receive 5 shorter code words than the values that occur frequently. Entropy encoded values are then supplied as real main information to bit stream formatter 1004, which then produces the encoded audio signal on the output side, according to a specific bit stream syntax.

As already illustrated, a finer quantizing step dimension is used in this iterative quantization, in the event that the interference introduced by a quantized step dimension is greater than the limit, this being done in the hope that it will lead to a reduction in noise of quantization due to the fact that the quantization performed is finer.

This concept is disadvantageous because, with the dimension of the finer quantizing step, the amount of data to be transmitted increases naturally, and thus, the compression gain decreases.

It is the object of the present invention to provide a concept for determining a quantizing step dimension which, on the one hand, introduces low quantization interference and provides, on the other hand, a high compression gain.

This object is achieved by means of equipment for determining a quantizing step dimension as in claim 1, by a quantizing step dimension method as in claim 8 or by a computer program as in claim 9

Ο

of the patent.

The present invention is based on the findings that an additional reduction in the interference power, on the one hand, and at the same time an increase, or at least the preservation of the coding gain, can be achieved when at least several coarser quantized step dimensions are attempted, even when the introduced interference is greater than a limit, instead of performing a finer quantization, as was done in the prior art. It turns out that even with coarse quantizing step dimensions, reductions in interference introduced by quantization can be obtained, to be exact in those cases where the dimension of the coarse quantizing step impacts the value to be quantized better than the dimension of the finer quantizing step . This effect is based on the fact that the quantization error depends not only on the dimension of the quantizing step, but naturally also on the values to be quantized. If the values to be quantized are very close to the step dimensions of the coarse quantizer step dimension, a reduction in the quantization noise will be obtained

0 while increasing the compression gain (since the quantization was more coarse).

The inventive concept is very profitable, particularly when there are well-estimated dimensions of quantizing steps for the first dimension of quantizing steps, on the basis of which the limit comparison is made. In a preferred embodiment of the present invention, it is therefore preferred to determine the first dimension of the quantizer step by means of a direct calculation based on the energy of the average noise rather than on the basis of a · · · · · · · · · · · • · · ·

............

·· · ♦ ·· • · · · ’· · * * * * ϊ worst case scenario. Thus, the cycuitps..dê..iteração? · * D ^ cfaccording to the prior art may already be considerably reduced or may become totally obsolete.

inventive post-processing of the dimension of the quantizing step will then be experienced, just once again, an even more coarse quantizing step dimension in the configuration, in order to benefit from the described effect of better impact of a value to be quantized. If, subsequently, the interference obtained by the coarser quantizing step 10 dimension is less than the previous interference or even less than the limit, more iterations can be performed to try an even coarser quantizer step dimension. This procedure of increasing the dimension of the quantizer step continues for a while until the introduced interference grows again. Then, a termination criterion is reached, so that the quantization is done with the dimension of the stored quantizer step that provided the least interference introduced, so that the coding procedure proceeds as necessary.

In an alternative configuration of the present invention, to estimate the size of the first quantizer step, an approximate synthesis analysis approach can be made as in the prior art, which continues for a while until a termination criterion is reached there. Then, inventive post-processing can be employed to eventually check whether it may be possible to achieve equally good interference results, or even better interference results with a coarser quantizing step dimension. If it is determined that a coarser quantizing step dimension • .igualVftéhêe •• teow ** oU even better with respect to the introduced interference, that dimension of. step will be used for quantification. If it is determined, however, that a coarse quantization does not produce positive effects, that dimension of the quantizing step originally determined will be used for a possible quantization, for example, by means of an analysis / synthesis method.

According to the invention, any quantizer step dimensions can be used to make a first comparison of limits. It is irrelevant if this first dimension of the quantizing step has already been determined by the analysis / synthesis schemes or even by means of direct calculations of the dimensions of the quantizing steps.

In an alternative configuration of the present invention, this concept can be used to quantize an audio signal present in the frequency range. However, this concept can also be used to quantify a time domain signal that comprises audio and / or video information.

In addition, it should be noted that the limit used to compare is a permissible psycho-acoustic or psycho-optic interference, or another limit to be reduced. For example, this limit may actually be a permissible interference provided by a psycho-acoustic model. However, this limit can also be a predetermined interference introduced for the dimension of the original quantizer step, or any other limit.

It should be noted that the quantized values do not

Μ) must necessarily be coded which can alternatively be encoded using another entropy coding, such as an arithmetic coding. Alternatively, the quantized values can also be encoded in a binary way, since this encoding also has the effect that, in order to transmit smaller values or values equal to zero, fewer bits are needed than are necessary for the transmission of higher values or , in general, non-zero values.

For the determination of the starting values, that is, the dimension of the quantizer step 1, the iterative approach can preferably be totally, or at least largely eliminated, if the dimension of the quantizer step is determined from a direct energy estimate. of noise. Calculating the dimension of the quantizer step from an accurate noise energy estimate is considerably faster than calculating an analysis by synthesis circuit, since the values for the calculation are directly present. It is not necessary to first make and compare several quantization attempts until a quantizing step dimension is found that is favorable for coding.

However, as the quantizer characteristic curve used is a non-linear characteristic curve, the non-linear characteristic curve must be taken into account when estimating noise energy. It is no longer possible to use the simple noise energy estimate for a linear quantizer, as it is not sufficiently accurate. According to the invention, a quantizer is used which has the following quantization characteristic curve:

• ♦ · · ·· • * · • · · ♦ • «y _{ - round

In the above equation, x _£ are the spectral values to be quantized. The starting values are characterized by yi, Yi and are therefore quantized spectral values. q is the dimension of the quantizer step. Round is the rounding function, which is preferably the nint function, nint meaning the nearest integer. The exponent that makes the quantizer a non-linear quantizer is called a, a being different from 1. Typically, the exponent a will be less than 1, so that the quantizer has a compression characteristic. With layer 3, and with AAC, the exponent α is equal to 0.75. The parameter s is a constant additive that can have any value, but can also be zero.

According to the invention, the following connection is used to calculate the dimension of the quantizer step.

ΣΙΜ

7 '

12th '

Σ-υ- ’

Being α equal to, the following equation results:

| l / 2

In these equations, the term on the left signifies the permissible THR interference in a frequency band and that is

0 provided by a psycho-acoustic module for a scale factor band with the frequency lines from i equal to ii to i equal to ₂ . The above equation allows an almost exact estimate of an interference introduced by a quantizer step dimension q for a non-linear quantizer that has the quantizer characteristic curve above with an α exponent other than 1, characterized

- 13 • ·· * · · '·· ·· ::

······ l · * · · · »• ·: ϊ ·. ·» ···:

by the fact that the nint function of ec ^ a cations. ^, 'qnant'izs'd®ra ^,, p2 * õ ^? orients the real quantization equation, which is rounded to the nearest integer.

It should be noted that, instead of the nint function, any desired rounding function can be used, for example, also rounding to the next even integer or the next odd integer, or rounding to the next integer of 10, etc. Generally speaking, the rounding function is responsible for mapping a value from a set of values that have a specific number of allowed values to a set of values that have a smaller second number of specific values.

In a preferred configuration of the present invention, the quantized spectral values have been previously subjected to TNS processing and, if dealing with, for example, stereo signals, for central / lateral encoding, provided that the channels are such that the central / lateral encoder has been activated.

Thus, the scaling factor for each scaling factor band can be directly indicated and can be fed into the respective audio encoder with the connection between the dimension of the quantizing step and the scaling factor, which is provided according to the following equation :

g - _

The scale factor results from the following equation.

«Scf = 8.8585 - [log _I0 (6.75 -THR) -log ₁₀ (FK4C)];

ΣΗ ¹ ' ² = ™ c

In a preferred configuration of this £ 2>

, ♦ · ♦ · · * ·;

• »· • * · • ··· invention, can also be made · * usç5» 'give *> i * úrna * * itWI16fÇSb ** post-processing based on a principle of analysis by synthesis, in order to vary the dimension a little the quantizer step, which was calculated directly without iteration, for each scale factor band in order to achieve the present ideal.

However, compared to the prior art, the already very accurate calculation of the starting values allows for a very small iteration, although it has happened that in the vast majority of cases, the downstream iteration can be totally dispensed with.

The preferred concept based on the calculation of the step dimension using the energy of the average noise thus provides a real and good estimate, since unlike the previous technique, it does not operate with the worst case scenario, but uses an expected value of the quantization error. as a basis and thus allows, with subjectively equivalent quality, a more efficient encoding of the data with a considerably reduced bit count. In addition, a considerably faster code can be obtained due to the fact that the iteration can be totally dispensed with and / or that the number of iteration steps can be clearly reduced. This is notable, in particular, because the iteration circuits in the prior art encoder were essential to the overall time requirement of the encoder. Thus, even a reduction of one or less iteration steps leads to considerable overall time savings for the encoder.

The preferred configurations of the present invention will be explained below in detail, with reference to the accompanying figures, characterized by the fact that:

diagram · * ·· cfe · ’’ block · · ·· tfe · · * a quantized audio signal;

Fig. 1 is a device for the determination of

Fig. 2 is a flow chart of the post-processing representation according to a preferred embodiment of the present invention;

Fig. 3 shows a block diagram of a prior art encoder according to the AAC standard;

Fig. 4 is a representation of the reduction of quantization interference by a coarser quantizing step dimension; and

Fig. 5 shows a block diagram of the equipment of the invention for determining a dimension of quantizing step for the quantization of a signal.

inventive concept will be presented below with reference to Fig. 5. Fig. 5 shows a schematic representation of an equipment for determining the dimension of the quantizer step for the quantization of a signal comprising audio or video information and being provided by means of a signal input 500. The signal is provided to a means 502 to provide a first dimension of the quantizer step (QSS) and to provide an interference limit which will also be referred to below as an introductory interference. It will be noted that the interference limit can be any limit. Preferably, however, it will be an optically introducing psycho-acoustic or psycho25 interference, this limit being selected so that a signal in which the interference has been introduced will still be perceived as not interfered by human observers or listeners.

The limit (THR), as well as the "quantitative step", are provided to a medium 504 for the determination of the first real interference introduced by the first dimension of the quantizing step. The determination of the interference actually introduced is made preferably by quantization, using the first dimension of the quantizing step, by re-quantizing using the first dimension of the quantizing step, and by calculating the distance between the original signal and the re-quantized signal. Preferably, when spectral values are being processed, the corresponding spectral values of the original signal and the re-quantized signal are squared, so as to then determine the difference of the squares. Alternative methods can be used to determine the distance.

The medium 504 provides a value for a first interference actually introduced by the first dimension of the quantizer step. This first interference is provided, along with the THR limit, to a 506 medium for comparison. The means 506 makes a comparison between the THR limit and the first interference actually introduced. If the first interference

0 actually introduced is greater than the limit, medium 506 will activate medium 50 8 to select a second dimension of quantizing step, medium 508 being configured to select a second dimension of coarse quantizing step, that is, greater than first dimension of quantizer stage. The second quantizer step dimension selected by means 508 is provided to means 510 to determine the second interference actually introduced. For this, the medium 510 obtains the original signal, as well as the second dimension of the quantizing stage and • · · · · · · * »· · · · · * * _φ · _φ · ·· J again performs a quantization using · · ^ •• sdgurfda · diftifâilScío ^ of a quantizing step, a re-quantization using the first dimension of a quantizing step, and a calculation of the distance between the requantized signal and the original signal, in order to provide a 512 5 means to compare with a measure of the second interference actually introduced. The comparison medium 512 compares the second interference actually introduced with the first interference actually introduced or with the THR limit. If the second interference actually introduced is less than the first interference actually introduced, or even less than the limit

THR, the second quantizer step dimension will be used to quantize the signal.

It will be noted that the concept shown in Fig. 5 is only schematic. Of course, it is not absolutely necessary to provide separate means of comparison to perform the comparisons in blocks 506 and 512, but it is also possible to provide a single adequately controlled means of comparison. The same applies to means 504 and 510 for determining the interferences actually introduced. These, too, need not necessarily be configured as separate means.

In addition, it will be noted that the quantizing medium does not necessarily need to be configured as a medium separate from the 510 medium. To be exact, the signals that are quantized by the second quantizing step dimension are typically generated as early as in the 510 medium, when the medium 510 performs quantization and re-quantization to determine the interference actually introduced. The quantized values obtained there can also be stored and released as a signal • · ”·· ·· · · ···. ········» · · quantized when the medium 512 for comparison was foifLetfe iim ** fè * ^ last positive, so that the means 514 for quantization merges, as such, with the means 510 to determine the second interference actually introduced.

In a preferred embodiment of the present invention, the THR limit is the highest introductory interference determined by means of psycho-acoustics, in which case the signal being an audio signal. Here, the THR limit is provided by a psycho-acoustic model that operates in a conventional manner and that provides, for each scale factor band, an estimated maximum quantization interference that can be introduced in that scale factor band. The maximum introducible interference is based on the masking limit where it is identical to the masking limit or derived from the masking limit, in the sense that, for example, 15 is coded with safe spacing so that the introductory interference is less than the masking limit, or that a more offensive code be made in order to obtain a reduction in the bit rate, specifically in the sense that the permissible interference exceeds the limit of

0 masking.

A preferred way of implementing medium 502 to provide a first dimension of quantizer step will be presented below, with reference to Fig. 1. In this aspect, the functionalities of medium 50 of Fig. 2 and medium 502 of Fig. 5 are 25 as same. Preferably, the means 502 is configured to have the features of the means 10 and the means 12 of Fig. 1. In addition, in this example, the quantizer 514 of Fig.

identical to the quantizer 14 of Fig. 1.

is configured to be ··· · * · ·· ··· ·· ···· · * ·· ·· • · ·· ·· · * ·· ·· ·· · • * ·· ···· · · · · · · "The· · · · ·· " · ···

It will also be presented e & ãwxõ · · with * reference · to Fig. 2, a complete procedure that, if the interference introduced exceeds the limit, will also try to obtain dimensions of coarse quantizer steps.

In addition, the left branch of Fig. 2, which shows the inventive concept, is expanded in the sense that in the event that the introduced interference exceeds the limit and that the increase in the dimension of the quantizer step does not produce any effect, and if the needs of the bit rates are not particularly strict and / or if there is still some space in the bit saving bank, an iteration is made using a smaller, that is, finer dimension of quantizer step.

Eventually, the effect on which the present invention is based will be presented below, with reference to Fig. 15 4, specifically the effect that, despite the increase in the dimension of the quantizer step, reduced noise can be obtained

quantization and, associated with this, an increase in gain in compression. THE Fig. 1 shows an equipment for The 20 determination of one quantized audio signal that is given as an

spectral representation in the form of spectral values. It will be noted, in particular, that in the event that - with reference to Fig. 3 - no TNS processing and no central / lateral coding have been done, the spectral values are directly the starting values of the filter bank. However, if only TNS processing is performed without central / lateral coding, the spectral values provided for the 1015 quantizer are residual spectral values, since they are formed from the TNS prediction filtering.

If TNS processing including central / lateral coding is employed, the spectral values sent to the inventive equipment are spectral values of a medium channel, or spectral values of a side channel.

To begin with, the present invention includes a means for providing permissible interference, indicated by 10 in Fig.

1. The psycho-acoustic model 1020 shown in Fig. 3, which is typically configured to provide interference or a

Limit and permits, also known as THR, for each scale factor band, that is, for a group of several spectral values that are spectrally adjacent to each other, can serve as a means of providing permissible interference. The allowable interference is based on the psycho-acoustic masking limit 15 and indicates the amount of energy that can be introduced into an original audio signal without the interference energy being perceived by the human ear. In other words, the permissible interference is the portion of the signal artificially introduced (by quantization) that is masked by the actual audio signal.

medium 10 is shown to calculate the permissible THR interference for a frequency band, preferably a scale factor band, and to supply that band to a downstream medium 12. The medium 12 serves to calculate a piece of the dimension information of the quantizer step for the frequency band for which the permissible THR interference was indicated. The medium 12 is configured to supply the dimension information piece of the quantizer step q for a medium a «t · · • * ·« · · 4 φ * * · · · · «« * · downstream 14 for quantization. The medium T4 pât * a *? 2uâhti ^; a $ ãó * 'dpéra' * according to the quantization specification in block 14, using the dimension information of the quantizer step in the case shown in Fig. 1, to initially divide a spectral value 5 Xi by the value of q, and then exponentiate the result with the exponent cx other than 1, and then add an additive factor s, as may be the case.

This result is then sent to a rounding function which, in the configuration shown in Fig. 1, selects the nearest integer. According to the definition, the integer can be generated again by cutting the digits behind the decimal point, that is, always rounding down. Alternatively, the nearest integer can also be generated by rounding down to 0.499 and rounding up to

0.5. Alternatively, the nearest integer can be determined by always rounding up, depending on the individual implementation. However, instead of the nint function, any other rounding function can be implemented that, in general, maps a value that must be rounded, starting from

0 from a first and largest set of values to a second, smaller set of values.

The quantized spectral value will then be present in the frequency band at the output of the medium 14. As can be seen in the equation shown in block 14, the medium 14 will also be naturally provided, in addition to the dimension of the quantizer step q, with the spectral value to be quantized in the contemplated frequency band.

It will be noted that the medium 12 does not need, · ··· · *, · »··· a ·· ··« ··· · · · · · · a · ·· · i · * · «·« · · · * · * * I «*« a ♦ · · · · · · · · • · necessarily calculate direct â € * dirflênáâo * cfã '' * 'êtà'pa quantizadora q, but this as information of the dimension of the stage alternative quantizer, the scale factor can also be calculated as used in audio encoders based on prior art transformation. The scale factor is linked to the dimension of the actual quantizing step by means of a relationship shown to the right of block 12 in Fig. 1. If the means for the calculation are still configured to calculate, such as information on the dimension of the quantizing step, the factor scale factor, this scale factor will be provided to medium 14 for quantization, which will then use, in block 14, the value of 2 ^{1/4 sce} for the quantization calculation instead of the q value.

A derivation of the form given in block 12 will be shown below.

As already shown, the quantizer exponentially as shown in block 14 obeys the following relationship:

y- ~ round - +5

The reverse operation will be presented as follows

This equation thus represents the necessary operation for re-quantization, characterized by the fact that Yi is a quantized spectral value and where x / is a re-quantized spectral value. Again, q is the dimension of the quantizer step that is associated with the scale factor through the relationship shown in

Fig. 1 to the right of block 12.

As expected, in the event that α is equal to 1, the result will be consistent with this equation.

3Ά

I »· · · · * · · ♦ # *» · -t · * • · · · · · · ·· · »··» · · · ·· * a »··« · · ·· ·

If the equation above is * & * orfiàdâ · a ** ufh VêVóF cf spectral values, the total noise power in a band indicated by index i will be given as follows:

ΣΜ

12α ·

In summary, the expected value of the quantization noise of a vector is determined by the dimension of the quantizer step q and the aforementioned form factor that describes the distribution of the quantities of the vector components.

The form factor, which is the rightmost term in the 10 equation above, depends on the actual input values and only needs to be calculated once, even if the above equation is calculated for desired THR interference levels to varying degrees.

As already seen, this equation with α equal to% is simplified as follows:

The left side of this equation is thus an estimate of the energy of the quantization noise which, in a limit case, is in accordance with the permissible energy of the noise (limit).

Therefore, the following approach will be taken:

Χ | Δχ ,. | ² = 7ϊΰ?

i

The sum of the roots of the frequency lines in the right part of the equation corresponds to a measure of the uniformity of the frequency lines and is known as the form factor preferably as soon as possible in the encoder:

Thus, the following results:

Σ | χ, Γ = FFAC, ·· ···· ·:: ·, · • ·· ··

.......

3/2

THR * ° --FFAC

6.75 q here corresponds to the dimension of the quantizer step. With AAC, it is specified as:

scf is the scale factor. If the scale factor is determined, the equation can be calculated as follows, based on the relationship between the step dimension and the scale factor:

₂ (3/8) _SC /

THR * ~ --FFAC

6.75

2 (3/8) «/ _

6.75-77¾

FFAC, 8 .6.75-77¾.

^ ^scf = i ^{los2 (} - ^^ ⁾ «scf = —1— [log ₁₀ (6.75 · THR) - log ₁₀ (FFAC)] scf = 8.8585 · [log ₁₀ (6.75 THR) - log ₁₀ (FK4C)]

The present invention thus provides a closer connection between the scf scale factors for a scale factor band that has a specific form factor and for which a specific THR interference limit is given, which typically originates from a psycho model -acoustic.

As already demonstrated, the calculation of the step dimension using a medium noise energy provides a better estimate, since the base used is the expected value of the quantization error, instead of a worst case scenario.

V

M ·· ···· • · · • · · · i nvêíífe ivo · * * é · âdé £ [tíá ão ‘para

Thus, the concept determines the dimension of the quantizer step and / or, in equivalence to this, the scale factor of a scale factor band without iterations.

However, post-processing as will be shown below via Fig. 2 can also be done if the calculation time requirements are not very strict. In a first step in Fig. 2, the first dimension of the quantizer step (step 50) is estimated. The estimation of the first dimension of the quantizer step (QSS) is done using the procedure shown in Fig. 1. Then, a quantization is made using the first dimension of the quantizer step in a step 52, preferably according to the quantizer as shown using block 14 of Fig. 1. In sequence, the values obtained with the first dimension of the quantizing step are re-quantized, in order to then calculate the introduced interference. Then, in a step 54, it is checked whether the introduced interference exceeds a predefined limit.

It will be indicated which step size

20 quantizer q (or scf) what was calculated through the connection represented on the block 12 is approximation. If the connection presented in block 12 : gives Fig. 1 was really exact, must be

established in block 54 that the introduced interference corresponds exactly to the limit. However, due to the nature of

As the connection approaches block 12 of Fig. 1, the interference introduced may exceed or fall below the THR limit.

In addition, it will be noted that the deviation from the limit will not be particularly large, not least because, nevertheless, «

i

4S

, .........

·· ·.

• · · · · ·. · ·· · · · · · · • ê slap · * £> 4 * · equ ê * * tíâ ãnclo a will be present. If it is determined, the first dimension of the quantizer step, the introduced interference falls below the limit, that is, if the question in the step is answered in the negative, the right branch of Fig. 3 will be chosen. If the introduced interference falls below the limit, this means that the estimate in block 12 of Fig. 1 was very pessimistic, so that in step 56, a dimension of the quantizer step that is coarser than the second quantizer step is established.

The degree to which the second dimension of the quantizer step is coarser, in comparison, than the first dimension of the quantizer step can be selected. However, it is preferable to adopt relatively small increments, since the estimate in block 50 will already be relatively accurate.

Using the second dimension of the coarse (larger) quantizing step, a quantization of the spectral values is made, a subsequent re-quantization and a calculation of the second interference that corresponds to the second dimension of the quantizing step in a step 58.

In a step (60), it is then verified whether the second interference, which corresponds to the second dimension of the quantizer step, still falls below the original limit. If this happens, the second dimension of the quantizer step is stored (62), and a new iteration is started in order to establish an even coarser dimension of the quantizer step in one step (56). Then, step 60 and, as the case may be, step 62, are then performed again using an even coarser dimension of the quantizer step, so as to start a new one again.

- 27 • · * · · · · · · · · · · · · ·?

iteration. If it is found, during Jíma .Lt'era? Ão'tia · eíâVá ”60 *, 'that the second interference does not fall below the limit, that is, it exceeds the limit, a termination criterion will have been reached, and after reaching the termination criteria, quantization (64) is performed using a dimension of the quantizer step that was last stored.

As the first dimension of the quantizing step was already a relatively good value, the number of iterations when compared with poorly estimated starting values will be reduced, which will lead to significant savings in calculation time when coding, since the iterations for calculating the dimension of the quantizer step consume most of the encoder calculation time.

Below will be shown an inventive procedure used when the introduced interference actually exceeds the limit, with reference to the left branch in Fig. 2.

Despite the fact that the interference introduced already exceeds the limit, an even coarser quantizing step dimension is established according to the invention (70), a quantization, a re-quantization and the calculation of the second noise interference, which corresponds to the second dimension of the quantizer step being carried out in step 72. Then, a check is made in step 74 of whether the second noise interference now falls below the limit. If so, the question in step 74 will be answered yes and the second dimension of the quantizer step will be stored (76). However, if it is found that the second noise interference exceeds the limit, quantization will be performed using the dimension of the quantizer step:::: ♦ ’<.

stored, or, if it has not been ^ rmazfeííTT ^ urrtet rtielHW's second dimension of the quantizing step, an iteration will be made, characterized by the fact that, as in the prior art, a dimension of the finer quantizing step is selected to push the introduced interference below the limit.

A discussion follows on why an improvement can still be obtained when an even coarser dimension of the quantizer step is used, particularly when the interference introduced exceeds the limit. Until now, it has always been operated on the assumption that a dimension of the finer quantizing step leads to less quantization energy introduced, and that a larger dimension of the quantizing step leads to greater quantization interference introduced. On average, this may be true, but it is not always true, and the opposite will be true, in particular, for even more finely populated scale factor bands and, in particular, when the quantizer has a nonlinear characteristic curve. It has been determined, according to the invention, that in some cases that should not be underestimated, a dimension of the coarser quantizing step leads to less interference introduced. This can lead back to the fact that there may also be a case when a dimension of the coarse quantizer step reaches a spectral value to be quantized better than a dimension of the finer quantizer step, as will be presented using the example below with reference to Fig 4.

As an example, Fig. 4 shows a quantization characteristic curve (60), which provides four quantization stages 0, 1, 2, 3, when the input signals between 0 and 1 are quantized. The quantized values correspond to 0.0; 0.25;

SP

0.5 and 0.75. In comparison, a curve: carafc> t * eríffit * re & de • CfllâfiE Different and more coarse design is drawn in dotted lines in the

Fig. 4 (62), which only has three stages of quantization that correspond to absolute values 0.0; 0.33; 0.66. Thus, in the first case, that is, with the quantizer characteristic curve 60, the dimension of the quantizer step is equal to 0.25, characterized by the fact that in the second case, that is, with the quantizer characteristic curve 62, the dimension of the quantizer step. quantizer step is equal to 0.33. The second quantizer characteristic curve (62), therefore, has a coarser dimension of the quantizer step than the first quantizer characteristic curve (60), which represents a fine quantization characteristic curve. If the value Xi = 0.33 to be quantized is taken into account, it can be seen from Fig. 4 that the quantization error using the four-stage fine quantizer is equal to the difference between 0.33 and 0.25, being therefore, 0.08. In contrast, the error in quantization using three stages is equal to zero, due to the fact that the quantizer stage reaches exactly the value to be quantized.

Therefore, it can be seen from Fig. 4, that a more coarse quantization can lead to a lesser quantization error than a fine quantization.

In addition, a more coarse quantization is the decisive factor for the need for a lower starting bit rate, since the possible states are only three, that is, 0,

1, 2, unlike the case of the finer quantizer, where four stages 0, 1, 2, 3 must be marked. In addition, the coarse quantizing step dimension has the advantage that more values tend to be quantized further away from 0 than with • · í:: ·· * .: · ·. . ·. * · ♦:

a dimension of the quantisation step ntais »í5isiâ *** c> rfde 'me * hò * s *' values are quantized further away from 0. Even so, when several spectral values are considered in a scale factor band, quantization to 0 leads to an increase in the quantization error, but this should not necessarily become problematic, since a dimension of the coarser quantizing step can to reach other more important spectral values in a more accurate way, so that the quantization error is canceled and even overcompensated by the coarser quantization of the other spectral values, at the same time occurring a lower bit rate.

In other words, the result of the achieved encoder is better, that is, as the inventive concept reaches a smaller number of states to be signaled and, at the same time, it provides a better impact of the quantization stages.

According to the invention, as shown in the left branch of Fig. 2, an even coarser dimension of the quantizer step is attempted, starting from estimated values (step 50 of Fig. 2), when the introduced interference exceeds the limit , in order to benefit from the effect represented using Fig. 4. In addition, it was determined that this effect is even more significant with non-linear quantizers than in the case, mentioned in Fig. 4, of two characteristic curves linear quantizers.

The presented concept of dimension of the post-processing quantizer step and / or post-processing scale factor thus serves to improve the result of the scale factor estimator.

.............. .......

·· ··· *. * ·· · • · · · ····. · · ··· ··· •• t.ií ·· · · of the dim $ nsõ53 '.. aa ·· efcapa ”qtiSMti * scaling factor generator (50 in Fig. 2), are analysis by synthesis, new dimensions of

Starting from determined in the estimator determined, in the quantizer stage step that are the largest possible and for which the error energy falls below the predefined limit value.

Therefore, the spectrum is quantized with the dimensions of the quantizing step calculated, and the energy of the error signal, that is, the sum of squares of differences in the original and quantized spectral values is preferably determined. Alternatively, for the determination of error, a corresponding time signal can also be used, although spectral values are preferred.

The dimension of the quantizer step and the error signal are stored with the best result obtained so far. If the calculated interference exceeds a threshold value, the following approach is taken:

scaling factor within a predefined range is varied around the value originally calculated, with the use, in particular, of coarser quantizing step dimensions (70).

For each new scale factor, the spectrum is again quantized, and the energy of the error signal is calculated. If the error signal is less than the smallest that has already been calculated, the current dimension of the quantizer step is considered, together with the energy of the associated error signal, as the best result obtained so far.

According to the invention, not only relatively small graduation factors, but also relatively large graduation factors s ^ ò * Levado & βπτ · 'ôtfntci, in order to benefit from the concept described with reference to Fig.

4, particularly when the quantizer is a non-linear quantizer.

However, if the calculated interference falls below the limit value, that is, if the estimate in step 50 was too pessimistic, the scale factor will vary within a predefined range, around the value originally calculated.

For each new scale factor, the spectrum is re-quantized, and the energy of the error signal is calculated.

If the error signal is less than the smallest that has been calculated until then, the current dimension of the quantizer step is considered, together with the energy of the associated error signal, as the best result obtained so far.

However, here we only take into account

factors in relatively coarse graduation, in order to reduce O number of bits required for coding spectrum in audio.

Depending on the circumstances, the inventive method

0 can be implemented in hardware or software. The implementation can be done in a digital storage medium, in particular a disk or CD with control signals with electronic reading, which can cooperate with a programmable computer system, so that the method is carried out.

In general, the invention thus consists of a computer program product equipped with a program code, stored in a machine-readable carrier, for carrying out the inventive method, when the

The computer operates on a computer. JEm © Utra’s- ^ ãi * ãvras ·, ·· SVlrffrenference can thus be performed as a computer program with a program code for carrying out the method, when the computer program is operated on a computer.

1/4

Claims (5)

1/5 ·· · • · · · · ···· · · · · »,« ··· • · · * * ··········· SPECTRAL VALUES QUANT1ZED IN THE FREQUENCY BAND Figl 1. Equipment for determining the dimension of the quantization step of a signal that comprises audio or video information, the equipment characterized by the fact that it comprises: means (502) to provide a first dimension of the quantizing step and an interference limit ; means (504) for determining a first interference introduced by the first quantizer step dimension; means (506) for comparing the interference introduced by the first dimension of the quantizing step with the interference limit; means (508) for selecting a second quantizing step dimension, which is greater than the first quantizing step dimension, if the first introduced interference exceeds the interference limit; means (510) for determining a second interference introduced by the second dimension of quantizer step; means (512) for comparing the second interference introduced with the interference limit or the first interference introduced; and means (514) for quantizing the signal with the second dimension of quantizing step, if the second interference introduced is less than the first interference introduced or is less than the interference limit. 2/5 Fig 2 2 (l-a) where the means (514) for quantization is configured to quantize according to the following equation: y. = round where Xi is a spectral value to be quantized, where q represents the information of the quantizing step dimension, where s is a data that differs or is equal to zero, where α is an exponent other than 1, where round is a function rounding that maps a value from a first largest range of values to a value within a second and smallest range of values, where Y | Ax, | i (THR) is the permissible interference and where i is an operating index for the spectral values in the frequency band. Petition 870180051055, of 06/14/2018, p. 8/10 2/4 2. Equipment according to claim 1, characterized by the fact that the signal is an audio signal and comprises spectral values of a spectral representation of the audio signal, and where the means (502) for the provision is configured as a model psycho-acoustic that calculates a permissible interference for the frequency band, based on a psycho-acoustic masking limit. Petition 870180051055, of 06/14/2018, p. 7/10 3/5 Fig 3 (PREVIOUS TECHNIQUE) 3/4 5. Equipment according to any of the preceding claims, characterized by the fact that the means (508) for the selection is further configured to select a larger dimension of the quantizing step when the introduced interference is less than the permissible interference. 6. Equipment according to any of the preceding claims, characterized by the fact that the means (502) for provisioning is configured to provide the first dimension of the quantizing step as a result of an analysis / synthesis determination. 7. Equipment according to any of the preceding claims, characterized by the fact that the means (508) for selection is configured to change a dimension of a quantizing step for a frequency band regardless of a dimension of a quantizing step for another frequency band. 8. Equipment according to any of the preceding claims, characterized by the fact that the means (502) for provisioning is configured to determine a first dimension of the quantizing step as a result of a previous iteration step with the thickening of the dimension of the quantizing step, and where the interference limit is an interference introduced in the previous iteration step to determine the first dimension of the quantizer step. 9. Method for determining the dimension of the quantization step of a signal comprising audio or video information, the method characterized by the fact that it comprises the steps of: providing (502) a first step dimension Petition 870180051055, of 06/14/2018, p. 9/10 Equipment according to claim 1 or 2, characterized in that the means (504) for determining the first introduced interference, or the means (510) for calculating the second introduced interference is configured to quantize using a dimension quantizer step, to re-quantize using the quantizer step dimension, and to calculate a distance between the re-quantized signal and the signal, in order to obtain the introduced interference. 4/5 - ERROR WITH QUANTIZATION OF 4 STAGES (QU. FINA) 0.33 - 0.25 = 0.08 - ERROR WITH QUANTIZATION OF 3 STAGES (QU. THICK) 0.33-0.33 = 0.00 Fig 4 • · · · · · · · · · · · · · · · · · · · · · · S1 * ·· ·· ··· »· ♦ • · · · ♦ · • · ····· · *« · · · · • · ♦ · ··. · * · * ·· »··· 4? · · · · · · · · · · · · · · · · · · · · 4/4 quantizer and an interference limit; determine in (504) a first interference introduced by the first dimension of the quantizer stage; compare (506) the interference introduced by the first dimension of the quantizing step with the interference limit; select (508) a second dimension of a quantizing step that is greater than the first dimension of a quantizing step, if the first interference introduced exceeds the interference limit; determine in (510) a second interference introduced by the second dimension of a quantizing step; compare (512) the second interference introduced with the interference limit or with the first interference introduced; quantize in (514) the signal with the second dimension of quantizing step, if the second interference introduced is less than the first interference introduced or is less than the interference limit. Petition 870180051055, of 06/14/2018, p. 10/10 4. Equipment according to any of the preceding claims, characterized by the fact that the means (502) for providing the first dimension of the quantizing step is configured to calculate the dimension of the quantizing step according to the following equation: 2a Υ | Δχ | ² «-í__Vx ^: ν '12a ² ' 5/5 QUANTIZED Fig 5